Skip to content

Evaluate CodeGuard with local models and constrained context windows #73

@thomas-bartlett

Description

@thomas-bartlett

Summary

Evaluate whether CodeGuard remains useful when used with local models, smaller context windows, or constrained token budgets.

Questions

  • Can a local model use CodeGuard guidance effectively for generation or review?
  • Which current delivery path is most practical under constraints: rule files, skills, MCP, flattened prompt, or codeguard-reviewer?
  • How does repo/task size affect whether the guidance is usable?
  • What evidence should a tester record so results are comparable?

First Pass

Run one small generation task and one review task with a local model setup. Record model, host/tool, delivery path, context limits, latency, and whether the output clearly used CodeGuard-specific guidance.

Output

A short comment summarizing what worked, what failed, and what should be tested next.

Metadata

Metadata

Assignees

No one assigned

    Labels

    enhancementNew feature or requesthelp wantedExtra attention is needed

    Type

    No type
    No fields configured for issues without a type.

    Projects

    No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions